Child Support Enforcement Software Problems: The 7 That Break Cases, and How to Avoid Them

Why does someone always propose rebuilding the certified core?

Because the request that reaches procurement is reasonable. The current system is old, the screens are unforgiving, caseworkers are re keying between it and three spreadsheets, and the honest description of the problem sounds like the system should be replaced. A firm hears that description and quotes a replacement, because that is the shape of the request.

What makes it different here is certification and history. A statewide IV-D system carries federal certification, and the genuinely hard part of a replacement is not writing the new distribution engine, it is proving that the new engine produces identical results across a decade of historic cases where assignment status changed, families moved on and off assistance, orders were modified retroactively, and payments were reversed after they had already been disbursed. That proof is the project, and it does not fit inside the budget anyone brings to it.

The fix: scope the surround, not the core. Order establishment against the state case registry, distribution and federal reporting stay where they are. Employer portal, parent self service, worker triage and interstate workspace, document assembly and management reporting against your own copy of the data are all real projects with real returns, and none of them require certification. If a firm quotes you a replacement of the certified system as a software project rather than a multi year programme with federal funding approval, treat that as disqualifying information rather than as a competitive price.

What goes wrong with reading data out of the system of record?

This is the assumption that quietly ruins schedules. Design proceeds on the belief that the certified system will expose case, order, payment and balance data on request. Then someone actually asks, and the answer is a nightly extract in a fixed format, produced by a batch window that occasionally slips, containing most of what you need and not the fields you built the screen around.

Stale data is survivable if the product is designed for it and dishonest if it is not. A parent looking at a balance that is eighteen hours old is fine when the screen says so and shows the as of time. The same screen without that label generates a call, because the parent paid yesterday and cannot see it.

The second data trap is identity. One person can be an obligor on one case, a custodial party on another, and a participant under two spellings because the intake happened in two counties. Any surround system that joins on name and date of birth will merge people who should not be merged, and in this domain that is not a data quality issue, it is a disclosure incident between two adults who may be in conflict.

The fix: settle data access in week one, before design. Get the actual extract specification, not a description of it. Design every screen to display its own currency and to name the system of record for any figure. Use the certified system's participant identifiers as the spine rather than deriving your own, and where you must match, require explicit confirmation with an audit record rather than an automatic merge.

Why do employer and federal interfaces break after launch?

Most collections arrive through income withholding, which means the agency is running a data operation with tens of thousands of employers, most of whom have exactly one employee under an order and no payroll specialist. The interfaces around that are where a working system starts failing quietly.

The failure patterns are consistent. An employer sends a remittance with an identifier that does not match any case because someone transcribed it from a paper notice. An employee changes jobs and the withholding simply stops, with no event anywhere in the system to say so, so the case looks compliant until arrears build. A bonus is paid and reported afterwards rather than before, which is the difference between a lump sum intercepted and a lump sum spent. Two orders on one employee exceed the withholding limit under the Consumer Credit Protection Act, the employer guesses at proration, and the split is wrong for months.

New hire directory and locate feeds fail differently. They keep working and stop being read, because the results land in a report rather than in a worker's queue.

The fix: design the employer channel around the questions employers actually ask, not around the form you send them. Let them look up an order, confirm employment, report a termination with a last known address and any known new employer, report an upcoming lump sum before it is paid, and see exactly what to remit and where, with proration calculated for them when multiple orders exceed the limit. Then monitor the flow rather than the interface: an employer who remitted every fortnight for two years and has not remitted for six weeks is an exception, and that exception is worth more than any new enforcement remedy.

What happens when the federal tax information and privacy gaps are not covered?

Two gaps produce most of the pain, and both are discovered late in projects that did not name them at the start.

The first is federal tax information. If any component touches tax refund offset data, IRS Publication 1075 controls apply to the hosting environment, the people with access, the logging and the physical controls. That is a security programme with its own review cycle, not a configuration setting, and discovering it during a security review after the design is finished moves the schedule by months. Teams often assume the boundary is narrow, then find that a case screen showing an offset payment has pulled the whole component inside it.

The second is disclosure between the parties. A case involves two adults whose interests conflict and who may have a protective order between them. A self service channel that shows a payment history including an address, an employer name or a county of residence has disclosed something with real safety consequences. This is not a permission flag on a field, it is a separate view model per role.

The fix: confirm the federal tax information boundary with your agency security officer before design starts, because the answer determines architecture rather than following it. Build custodial and non custodial views as distinct products over the same case, so the default is that a field is absent unless deliberately included. Then have someone who understands domestic violence casework review the screens before launch, not after, and treat identity proofing as a design problem rather than a login step.

Should you build custom or configure what you already own?

Ask your incumbent first, and ask specifically. Systems and Methods Incorporated, Conduent and Tyler Technologies all offer portal and self service modules alongside the certified system, and where those modules do what you need, configuring them is faster, cheaper and carries no integration risk. That is a genuine answer for agencies whose main complaint is that parents cannot see a balance online.

Configuration stops being enough when the requirement is operational rather than informational. Employer workflows that calculate proration and accept a termination with a new employer, worker triage that shows what is blocking an interstate case and in whose court, document assembly for your state's exact forms with their conditional logic, and management reporting against your own copy of the data are all outside what these modules are shaped for.

Build the surround when caseworkers are re keying between the certified system and spreadsheets, when employers call because your notices are ambiguous, when your interstate cases lose weeks waiting on something the other state has not sent, or when your federal performance measures are being dragged down by workload rather than by law. Build a full replacement only where a county programme runs something genuinely unsupported, and plan that as a multi year programme with federal funding approval rather than as a software project.

How do hidden costs get into the quote?

Document assembly priced per document. Every form your state uses carries conditional logic and a court that rejects it if the layout is wrong, and the exceptions are what take the time.

Identity proofing treated as a login feature. You are exposing financial detail about a case involving two adults in conflict, so the proofing standard, the recovery path and the lockout policy are product decisions with legal review attached.

County versus state administration. In county administered programmes each local variation is a real branch in the workflow, and a quote built for one county rarely survives contact with the fourth.

Accessibility, language and the security review. Public facing government systems carry accessibility obligations and several languages, and environment approval, staff background requirements and logging design all consume calendar time regardless of engineering pace. All of it belongs in the plan from week one, because retrofitting any of it after design is finished costs more than building it in.

From Digital Heroes delivery experience, a well scoped component in this domain runs $150,000 to $450,000 over 5 to 9 months. That covers an employer portal, or a parent self service channel, or a worker triage and interstate workspace, or a document assembly engine, not all four.

What separates a build that works from one that fails here?

One number, one source. Every figure on every screen either comes from the certified system or is clearly labelled as derived, with the source and the as of time visible. Agencies that get this right stop having reconciliation conversations. Agencies that get it wrong add a second thing for staff to explain.

An event log where you do hold financial detail. Where a component legitimately tracks accruals, credits or adjustments, model the balance as a replay of dated events rather than as a stored number that gets updated. That is what lets you produce a line by line schedule for a hearing in seconds, and what makes a retroactive modification recalculate instead of requiring a manual write off somebody has to justify.

Reversal handled properly. Ask any prospective developer what happens when a payment is returned three weeks after it distributed and disbursed across two cases. The answer should involve reversing and re running, never journalling a correction. This single question sorts candidates faster than any other in the domain.

Deadlines modelled as first class objects. Interstate and enforcement actions carry clocks, and a workspace that shows what is blocking, in whose court and for how long moves the federal performance measures more than any new remedy, because the cases losing time are not hard cases, they are cases nobody looked at this month.

Ownership in writing before kickoff. The agency should own the repository, the data and the environments, with an unrestricted right to move the work. At Digital Heroes that is the default from the first commit, and in a domain this sensitive it should not be negotiable anywhere.